Putting Politics First: The Impact of Politics on American Religious and Secular Orientations. David E. Campbell University of Notre Dame

Transcription

1 Putting Politics First: The Impact of Politics on American Religious and Secular Orientations David E. Campbell University of Notre Dame Geoffrey C. Layman University of Notre Dame John C. Green University of Akron Nathanael G. Sumaktoyo University of Notre Dame Abstract Nearly all research on the political impact of Americans religious and secular orientations assumes that such orientations are exogenous to politics. Using multiwave panel and experimental data, we find that religious and secular orientations are endogenous to political orientations. In other words, religion and secularism are a consequence as well as a cause of politics. In showing this, we make three major contributions. First, we conceptualize and measure secular orientations in a new way not just as the absence of religion, but as an affirmative secular identity and positive commitment to secular principles. Second, our panel and experimental data allow for the most definitive test to date of whether political orientations exert a causal effect on religious and secular orientations. Third, we isolate the conditions under which politics affects religious-secular perspectives, thus identifying the mechanism that underlies political orientations. Word Count: 9070 _ The data, code, and any additional materials required to replicate all analyses in this article are available on the American Journal of Political Science Dataverse within the Harvard Dataverse Network, at doi: /dvn/ym4ldg This research was supported by the National Science Foundation under Grant No. SES Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. We are grateful for the research assistance of Patrick Schoettmer and Jeremy Castle. Thanks also to Christ the King Lutheran Church of South Bend, Indiana for the use of their facilities, as well as Todd Adkins and Dan Hubert for appearing in the experiment s news stories.

2 The increasing polarization of American politics has been accompanied by increasing sociodemographic differences between partisan and ideological camps (Mason 2016). As the link between sociodemographic orientations and politics has grown, scholars have theorized that causality runs in both directions. People not only choose political sides based on their sociodemographic profiles, but also may base some social preferences including their feelings about non-political social groups (Iyengar and Westwood 2015), where they want to live (Bishop 2009), and even who they want to date or marry (Iyengar, Sood, and Lelkes 2012) on their political profiles. At the forefront of such speculation have been Americans religious and secular orientations. Party coalitions have grown increasingly divided along religious lines, with the GOP becoming more religious and the Democrats growing more secular and non-religious (D Antonio, Tuch, and Baker 2013). Scholars typically have assumed that this expanding divide has resulted from individuals choosing a partisan side based on their religious and secular orientations (e.g. Layman 2001). However, recent work suggests that people also abandon religion or become more religiously devout based on their partisan and ideological tendencies (Patrikios 2008; Putnam and Campbell 2010). In fact, a leading explanation for the recent surge in the percentage of Americans claiming no religious affiliation is politics: liberals and Democrats rejecting organized religion as traditionalist religion becomes increasingly associated with conservatism and the Republican Party (Hout and Fischer 2002, 2014). 1

3 Here, we provide the most thorough examination to date of the degree to which secular and religious orientations are endogenous to political orientations. In doing so, we make three advances in the study of how politics is related to religion and secularism. First, we provide a new way of conceptualizing and measuring secularism. We recognize that secularism is not simply the absence of religion, but also entails active secularism an affirmative secular identity and commitment to secular principles. Second, most research relies on cross-sectional data, making it impossible to assess causal direction. Other work has employed two-wave panels that allow evaluation of causal claims, but cannot distinguish true change in religious proclivities from change due to survey measurement error. By employing both an original survey experiment and a multi-wave panel survey, we provide a more definitive test of whether citizens political tendencies effect change in their religious and secular tendencies. Third, we isolate the conditions under which political identities are most likely to affect religious and secular orientations, thus providing a direct test of the causal mechanism underpinning the links between politics and both religion and secularism. Our analysis reveals a clear reciprocal relationship between political and religious-secular orientations and shows that the impact of politics is strongest when citizens perceive an infusion of religion into politics generally, and the Republican Party specifically. Strikingly, the effect of political orientations on religious and secular characteristics is often stronger than the reverse effect. This lends credence to the politicized religion explanation for growing secularism. It 2

4 also offers support for the idea that political identities and attitudes are themselves important social identities, capable of shaping and changing other social identities. Religious and Secular Orientations in American Politics Over the last forty years, the religious divide between the Democratic and Republican parties has expanded. Religious people, especially traditionalists, have become more Republican while the Democratic coalition has grown less religious and more secular (Layman 2001; Green 2007; Claassen 2015) Importantly, the mass public recognizes this pattern. Campbell, Green, and Layman (2011) report that Americans strongly associate evangelical Christians with the GOP, tend to see religious people as mostly Republicans, and view non-religious people as mainly Democrats. Following the conventional assumption that citizens religious orientations are more deep-rooted than their political preferences, virtually all work on this development assumes that religious and secular orientations are the causal mover people choose or change their political attitudes based on their religion or secularism. However, a few studies suggest the reverse that politics can affect religion. Hout and Fischer (2002) first proposed this reversal as they sought an explanation for the growth of the religiously non-affiliated population (the Nones ). Noting that the rise of the Nones coincides with the burgeoning influence of traditionalist Christians in conservative and Republican politics, they argued that the rise resulted primarily from political moderates and liberals rejecting religious identity as a 3

5 negative reaction to the mixture of religion and conservative politics. This proposed malleability of religious identification is consistent with research showing that many Nones fluctuate between claiming and not claiming a religious affiliation (Lim, MacGregor, and Putnam 2010). Other work confirms that religion is endogenous to politics (Hout and Fischer 2014; Putnam and Campbell 2010) and shows that, besides Democrats and liberals becoming more secular, Republicans and conservatives are growing more religious (Patrikios 2008). While these studies demonstrate that politics can affect religious characteristics, they have not directly tested the proposed causal mechanism that religion s injection into American politics and its association with the Republican Party have led citizens to base their religious and secular orientations on their political orientations. Here, we specify and test a theoretical explanation for why political identities affect religious-secular orientations. Because political orientations such as partisanship, ideology, and cultural attitudes serve as important social identities, they may shape not only political preferences, but also social and religious perspectives. This may produce cognitive dissonance among Democrats and liberals with a religious identity, leading them to become Nones. It also may spur an increasing returns process in which Republicans and conservatives grow more deeply religious and Democrats and liberals become more committed to secularism. Because the growth of secular orientations is an important result of this process, it is important that we accurately conceptualize and measure secularism. Social scientists generally treat secularism as the absence of religiosity (Hansen 4

6 2011). 1 The recent focus on the rise of the Nones is the quintessential example, as Nones are defined as people without a religious affiliation. Other measures of secularism include disbelief in God, lack of religious attendance, and the nonsalience of religion. Because this definition encompasses only the absence of religion and not an active embrace of secular perspectives, we call it passive secularism, but we measure it as the inverse of a traditional index of religiosity. While identification as a None is a manifestation of being passively secular, as explained below the literature suggests a distinct theoretical expectation for it. Accordingly, we treat being a None separately from the other measures of passive secularism. We argue that just as religion is multi-dimensional (Kellstedt et al. 1996), so is secularism. Many secularists do not simply reject religion; they actively promote secular beliefs, such as the efficacy of reason and science, and human experience as a proper basis for ethical judgments. Moreover, to be actively secular does not preclude also being religious in some way. That is, someone can embrace a secular perspective while maintaining a religious identity and participating in religious activities. 2 This is not a possibility when secularity is defined only as non-religion, 1 A notable exception is Beard et al. (2013), who note that secularism is more than the absence of religion, although their article does not directly measure secular identify or belief. 2 In other work (Layman et al. n.d.), we develop the concept of active secularism more fully and examine its relationship with passive secularism and with political orientations. Most people who are high in active secularism are also high in passive 5

7 making it important to distinguish passive secularism, or the absence of religiosity, from active secularism, or the affirmation of secular identity and beliefs. A Theory of Politicized Religion We argue that not only are voters political outlooks shaped by their religious-secular worldview, but their religious-secular orientations also are shaped by their political perspectives. Moreover, the religious impact of political identity is not limited to identification as a None, but extends to other components of passive and active secularism. Our starting point is a social identity conceptualization of political identification. If partisanship is a deep-rooted social identity as a longstanding perspective on party identification contends (Campbell 1960; Green, Palmquist, and Schickler 2002) then it may shape social identities and preferences. Group membership encourages individuals to conform to group norms and behavior patterns and to differentiate themselves from opposing groups. Since the public views the GOP as the party of religion and the Democrats as the party of the secular, partisanship may spur Republicans to grow even more religious and Democrats to become less religious and more actively secular. We might expect a similar dynamic with ideology and attitudes on cultural issues. While ideology is thought of as an overarching judgment about the proper role of government, Conover and Feldman (1981) find that ideological identification secularism, but nearly 10 percent of active secularists have high levels of religiosity. See the Supporting Information for more details. 6

8 is based more on affect toward ideological groups and the social groups associated with them. Cultural attitudes, meanwhile, differ from most policy opinions, which typically are weak and unstable (e.g. Converse 1964). Not only are they closely connected to religion and secularism, but attitudes on cultural issues such as abortion and gay rights are more emotional and entrenched than other issue positions (Converse and Markus 1979). In fact, Goren and Chapp (forthcoming) claim that views on cultural issues are strong attitudes, rivaling party identification and religion as long-term predispositions and exerting a causal influence on both partisanship and religiosity. This is consistent with Killian and Wilcox (2008), who showed that abortion attitudes can lead people to change their party ID, as pro-life Democrats and pro-choice Republicans switch parties. In short, ideological identification and cultural attitudes also may represent social identities capable of affecting citizens' religious and secular proclivities. How should politically-driven change in religious and secular orientations occur? First, religion s association with the GOP should create cognitive dissonance (Festinger 1957) among religious Democrats and liberals. To resolve the dissonance, these citizens might change either their political ties or their religious identity (Margolis forthcoming). The conventional expectation would be political change, shifting people in a conservative, Republican direction. In contrast, Hout and Fischer (2002) argue that cognitive dissonance can be resolved by changing one s religious identity, or, specifically, abandoning religious identity. We further expect that the effect of political identities on religious and secular orientations is not limited to a shift in identity, but extends to members of 7

9 opposing political camps growing more polarized in their religious-secular proclivities. As the image of the GOP as religious and the Democratic Party as nonreligious and secular sharpens, Republicans may respond by becoming even more religious and Democrats may grow more actively secular deepening the religious and secular images of the parties and resulting in an increasing returns process (e.g. Pierson 2000). More-religious people should be drawn toward the Republican Party and more-secular people should be attracted to the Democrats, further increasing the pressure for Republicans to become more traditionally religious and for Democrats to be more actively secular. Similar processes should take shape for ideology and cultural attitudes. These dissonance and increasing returns processes have developed because of a particular set of conditions: an increased injection of religion into politics and especially a growing connection between religiosity and the GOP. Thus, the impact of political identities on secular and religious orientations should be strongest for citizens who perceive high levels of religious infusion in American politics and see a close connection between religious traditionalists and the Republican Party. This discussion points to three hypotheses: Dissonance Hypothesis The association of religion with the GOP should make Democrats and liberals more likely to identify as Nones. Polarization Hypothesis Over time, political identities should have a mutually reinforcing relationship with passive and active secularism. More-religious people should grow more 8

10 Republican and conservative while more-secular people become more Democratic and liberal. Republicans and conservatives should grow more religious while Democrats and liberals become less religious and more secular. Perceptual Hypothesis Both dissonance and polarization should be more pronounced among voters who perceive an association between religion and politics in general and between religion and the Republican Party specifically. Data and Measures We employ two original datasets. One is the Secular America Study (SAS), a four-wave panel survey conducted online by GfK (formerly Knowledge Networks). We conducted the first wave of the study in the fall of 2010, interviewing 2,635 respondents; 3 the second wave in the summer of 2011 (1,909 3 To construct the sample for the SAS survey, 4,450 panelists were randomly drawn from the GfK respondent panel. A total of 2,635 panelists responded to the invitation, yielding a final stage completion rate of 59.2%. The recruitment rate for this study, reported by GfK, was 16.3% and the profile rate was 64.3%, for a cumulative response rate of 6.2%. While regrettably very low, this rate is consistent with other studies using online panels. The SAS panel included a general population sample, an oversample of individuals with no religious affiliation, and an oversample of people aged 18 to 29. All of the analyses presented here are weighted to 9

11 panel respondents); the third wave in February and March 2012, (1,541 respondents); and the final wave in October and November 2012 (1,412 respondents). 4 Because our core indicators of active secularism did not appear until the second wave, our analysis draws on waves 2-4. Our second dataset is a survey experiment administered by GfK to a nationally representative sample of 1,023 subjects in February of Our hypotheses require measures of identification as a None, other aspects of passive secularism (i.e. the inverse of religiosity), and active secularism. We operationalize Nones as respondents who identify their religious affiliation as nothing in particular. 5 Because we define passive secularism as the absence of religion, we employ standard measures of religiosity: religious service attendance, account for the oversamples, as well as any demographic imbalance in the general population sample. 4 Panel attrition between waves 1 and 4 is 46 percent. Because this attrition rate is relatively high, our results should be viewed with caution. However, we are given more confidence because the demographic profile of the sample remains virtually unchanged across panel waves. The Supporting Information includes the rates of panel attrition and respondents demographic profile in each panel wave. 5 Respondents who refused to answer the religious affiliation question in either the pre- or post-test were coded as missing. Nones do not include respondents who identified as atheist or agnostic because these are included in the secular identity measure described below. 10

12 frequency of prayer, religious salience, strength of belief in God, and literal belief in the Bible all coded so that higher values reflect less religiosity. 6 Just as religiosity includes religious beliefs, devotion, and identity, active secularism encompasses secular beliefs, secular identity, and commitment to secular perspectives. We measure secular beliefs with a battery of questions we designed to gauge commitment to a secular worldview. Respondents indicated how much they agree with five statements about scientific evidence and human reason as the proper foundation for explaining natural phenomena, understanding human behavior, and defining moral parameters. Because active secularism is neither the absence of religiosity nor antagonism toward religion, none of the statements explicitly reference religion so that secularism and religiosity are not pitted against each other. Three of the statements are worded in a direction that affirms secular perspectives: (1) Factual evidence from the natural world is the source of true beliefs. (2) The great works of philosophy and science are the best source of truth, wisdom, and ethics. (3) To understand the world, we must free our minds from old traditions and beliefs. 6 The Supporting Information includes the exact wording of all questions used in this analysis. 11

13 To guard against response acquiescence, the other two statements are worded in a non-secular direction: (4) It is hard to live a good life based on reason and facts alone. (5) Values are more important than factual evidence in making moral decisions. To capture secularism s salience, we created a non-religious guidance question that parallels the standard question about religious guidance. It asks how much guidance respondents receive from non-religious beliefs, such as derived from science or philosophy. To tap into secular identity, we asked respondents to select which (if any) terms from a list of religious and secular identities describe them. The terms were ecumenical, mainline, charismatic/ Pentecostal, humanist, non-traditional believer, secular, atheist, fundamentalist, born again/ evangelical, agnostic, and spiritual, but not religious. Exploratory factor analyses of identification with these labels find clear evidence that the secular, humanist, atheist, and agnostic labels form a single dimension. 7 We measure secular identity as the number of these secular labels selected, which ranges from zero to three. 8 7 See the Supporting Information for the results. 8 Respondents could identify themselves as atheists or agnostics both in the religious or secular labels and in the religious affiliation question. Our secular identity count variable includes respondents identifying as atheist or agnostic for 12

14 To see if active secularism and passive secularism are indeed distinct dimensions, we undertake a confirmatory factor analysis of their indicators in wave 2 of the SAS panel (the first wave in which our secular beliefs items appear). We show the results in Table 1. 9 [Table 1 Here] We compare the fit to the data of two models one with all the active and passive secularism variables loading on a single factor, the other with our active secularism indicators loading on one factor and our indicators of passive secularism loading on a second factor. 10 Both models correct for measurement error in the observed indicators. They treat non-religious guidance, secular identity, and all of either question. As very few respondents chose all four secular labels (less than 10 per wave), we combined respondents choosing either three or four secular labels. 9 The Supporting Information includes the results of an exploratory factor analysis of the indicators of passive and active secularism, which parallel the results in Table All of our confirmatory factor and structural equation models were estimated with Mplus 7.31, using full information maximum likelihood estimation with robust standard errors ( MLR estimation in Mplus) and applying sampling weights. This produces estimates for all observations in the sample, even those with missing values on the variables in the model (unless an observation is missing on all of the observed endogenous variables). 13

15 the indicators of passive secularism as having random measurement error. 11 However, following Green and Citrin (1994), who note that survey batteries that include statements worded in opposite directions may produce non-random (correlated) measurement error, we allow the measurement errors for each of the secular belief indicators to be correlated with each other. 12 Although all of the factor loadings are statistically significant, 13 the active secularism measures load far more strongly on their own factor in the two-factor model than they do in the single-factor model. This suggests that their underlying orientation is distinct from passive secularism. 11 The confirmatory factor models follow standard assumptions for measurement models (Bollen 1989). The covariances between the measurement errors and the latent variables and between the measurement errors and the structural disturbance terms are set to zero, and the factor loading for one observed indicator of each latent variable is set to one (worship attendance and the factual evidence statement). 12 See the Supporting Information for further details. Following Green and Citrin (1994), we constrain all of the correlations between measurement errors to be equal, estimating a single error covariance parameter for all five of our secular belief indicators. 13 Throughout this paper, a statistically significant effect or loading is one for which a two-tailed test of significance falls at p=.05 or below. 14

16 That is confirmed by the goodness-of-fit statistics for the two models. The two-factor model has a smaller value than the one-factor model of the chi-square test of overall model fit, a smaller value of the root mean square error of approximation (RMSEA), and a larger value of the comparative fit index (CFI). An appropriate test of whether the difference in fit is statistically significant is the difference in the chi-square values for the two models this difference is overwhelmingly significant (p<.0001). 14 In short, active and passive secularism represent separate dimensions of secular orientations. 15 Assessing Dissonance: The Candidate Religion Experiment We first test the dissonance hypothesis with a survey experiment, conducted online in February A nationally representative sample was administered a pre-test to measure their religious identity and their degree of passive and active secularism. Roughly one week later they read a fictional newspaper story about a 14 The chi-square statistics for overall model fit that the MLR estimator in Mplus produces are scaled to make them robust to non-normality and non-independence of observations. This makes it necessary to conduct the scaled difference in chisquare test suggested by Satorra and Bentler (1994), as shown in Table For our experimental analysis, we construct the measures of active and passive secularism through factor analyses with the survey data in which our experiment appeared (see the Supporting Information). For our panel analysis, the measures are created through the confirmatory factor models included in our structural equation models. 15

17 congressional race in a nearby state. Respondents were randomly assigned to read a version of the story that varied the amount of religious rhetoric, endorsements, and imagery associated with the Democratic and Republican candidates ranging from none to moderate to high use of religion. With two candidates and three possible conditions each, there are nine versions of the story. 16 While the community, Summerville, and the newspaper, The Summerville Gazette, are fictional, the article was designed to look as realistic as possible. Upon reading the story, subjects answered questions to replicate the pre-test. This design enables a clean test of whether exposure to religious politics in the news story triggers a change in either identity or attitudes. The article features an open-seat congressional race between Democratic and Republican candidates who are demographically similar. The story is nonsensational, highlighting the candidates backgrounds and positions on key issues. In the control condition, neither candidate mentions religion; the treatments add religious references to the text used in the control. To reflect the current state of religion s deployment in electoral politics, the religious references are Christian in nature. Although neither candidate is identified as belonging to a particular religious denomination, the religious cues have an evangelical Protestant tinge consistent with the religious rhetoric typical in contemporary politics. Each story contains three essential elements that vary according to the treatment s dose of religious politics: text, endorsements, and photos. Table 2 displays the variation across treatments. 16 The Supporting Information displays all the treatments. 16

18 [Table 2 Here] The articles thus provide multiple cues that one or both of the candidates are engaged in religious politics. They simulate religious references by contemporary politicians, while holding constant everything else about the campaign. Given the frequency of religious references in politics, we undoubtedly are not exposing subjects to religious politics for the first time. Rather, the stories reflect religious cues that they have likely encountered before, bringing them to the top of subjects heads. If the dissonance hypothesis is correct, then when Democrats are exposed to a Republican candidate who is associated with religion, they will become more likely to identify as Nones. Lest it seem that identification as a None is unlikely to change on the basis of a single news story, previous research shows selfclassification as None to be unstable. Over the course of panel surveys, substantial percentages of people (about one-third over a one-year panel, nearly half over five years) switch back and forth between identifying as None and claiming a religious affiliation (Lim, MacGregor, and Putnam 2010; Putnam and Campbell 2010). Many Nones are thus liminals who may or may not think of themselves as having a religious identity depending on the context. We do not expect that one news story on a congressional campaign will move people from highly religious to highly secular. Rather, in keeping with Hout and Fischer s conceptualization of Nones as on the boundary between religious affiliation and non-affiliation, it might nudge some liminals into self-identification as a None. 17

19 To test the dissonance hypothesis, we regress identification as None on a set of dichotomous variables representing each treatment (the control condition is the baseline). Since the model controls for identification as None in the pre-test, a positive coefficient reflects a treatment s effect on the increase in the incidence of identifying as a None. Because random assignment to the treatments was successful, no control variables are necessary. 17 We use logistic regression, as the dependent variable is dichotomous. The first column of Table 3 contains the results for all respondents. As expected, exposure to a religious Democratic candidate produces no effects. However, we uncover the expected effects for two treatments in which the Republican is heavily associated with religion. We see positive and statistically significant coefficients for the treatments with high religion for the Republican and either no religion or only moderate religion for the Democrat. There are no effects for the Republican-moderate treatments, or when both candidates are heavily associated with religion. This suggests that dissonance may be triggered only by overt references to religion by a Republican when not counter-balanced by a similarly strong dose of religion from the Democrat. [Table 3 Here] 17 We tested the randomization with Tukey s Honestly Significant Difference tests. Across our treatments, there are no differences in education, race (white, black, Hispanic), gender, or party identification. 18

20 While finding an effect in the whole sample is informative, the dissonance hypothesis centers on effects for Democrats. Accordingly, the second and third columns of the table display the results for Democrats and Republicans. 18 Among Democrats there is a large and significant effect for exposure to the combination of the high-religion Republican and the no-religion Democrat, and a similarly sized effect for the high-religion Republican and the moderate-religion Democrat, although the latter fails to achieve conventional significance (p=0.15). Democrats appear to experience dissonance when the Republican s association with religion is not offset by religious references on the Democratic side. 19 As expected, there are no effects among Republicans. When the logit coefficients are converted to probabilities, we find that for the population as a whole, there is a.057 increase in the probability of identifying as None, while for Democrats only it is nearly twice as large a jump of.106. In both cases, the 95 percent confidence interval does not cross zero. 20 Movement in religious self-identification merely from exposure to a single newspaper story not only supports the dissonance hypothesis but confirms that 18 Republicans and Democrats include independents who lean toward one party or the other. 19 Because of the small cell sizes for Independents (mean of 19 cases per cell), we only present the results in the Supporting Information. 20 Confidence intervals are estimated with the margin command in Stata. See the Supporting Information for a figure displaying the experimental results. 19

21 religious affiliation is more malleable than typically recognized. Providing a concentrated dose of the sort of stimuli experienced by voters as they observe the political process apparently has triggered the dissonance that leads Democrats to disclaim a religious affiliation. 21 Reciprocal Effects of Secular and Political Orientations in the SAS Panel Do political orientations produce changes in religious-secular orientations or vice versa in the real world, outside of an experimental setting? To assess this, we turn to the SAS panel, first as a second test of the dissonance hypothesis, and then to test the polarization and perceptual hypotheses. We use a similar model to test all three hypotheses. The political variables in the models are party identification (a seven-point scale ranging from strong Republican to strong Democrat), ideological identification (a sliding scale ranging from extremely conservative to extremely liberal ), and cultural issue attitudes. Our measure of cultural attitudes combines views on abortion (a four-point scale ranging from never allow to always allow ) and views on how the law should define marriage (a sliding scale ranging from only as a union between one man and one woman to as a union between two people regardless of their gender ). 21 Given that the polarization hypothesis involves change in passive and active secularism, which should be more deep-seated orientations than self-identification as None, we would not expect the experimental treatment to have an effect on either. It does not. See the Supporting Information for details. 20

22 We evaluate the reciprocal relationship between political orientations and secular orientations by using waves 2-4 of our SAS panel to estimate structural equation models of the following form: (1) Secular Orientation it = α 1 + λ 1Secular Orientation i,t-1 + β 1Political Orientation i,t-1 + ε 1,it (2) Political Orientation it = α 2 + λ 2Political Orientation i,t-1 + β 2Secular Orientation i,t-1 + ε 2,it This model tests cross-lagged effects between a particular political orientation and one of our secular orientations, with both variables being shaped by their own past values and the past value of the other variable. The λ parameters connecting each factor at time t to its own value at time t-1 capture the expected individual-level stability in secular and political orientations over time. The parameter β1 linking political tendency at t-1 to secular orientation at t captures the potential influence of previously held political perspectives on current secular orientation. Because equation (1) already controls for the effect of previous secular disposition, β1 measures the impact of political orientation on change in secular orientations from t-1 to t. Similarly, the parameter β2 in equation (2) captures the influence of secular orientation on change in political perspectives. To ensure that any impact of political orientations and secular orientations on each other represent effects on actual change, we correct for measurement error in the observed indicators of political and secular orientations. Our models combine the structural model of cross-lagged effects with a measurement model (i.e. a confirmatory factor model) in which observed indicators are structured by both 21

23 latent true variables (e.g. active secularism) and measurement errors. For the measurement models to be identified, we need either multiple indicators of the latent variable or three or more waves of panel data (Bollen 1989). For active secularism, passive secularism, and cultural issue attitudes, we have both three waves and multiple observed indicators. The three waves of data also allow us to correct for measurement error in partisanship, ideological identification, and identification as None even with only one observed indicator of each variable. In addition to the standard measurement model constraints (see footnote 11), these single-indicator measurement models require additional restrictions for identification (Wiley and Wiley 1970). We assume that the measurement errors of the observed indicators are uncorrelated across panel waves and that the effects of latent variables on the single observed indicator are equal to one. For the multipleindicator latent variables, we allow the measurement errors to be correlated across panel waves and estimate all factor loadings except one per latent variable. We also place constraints on some of the structural parameters. We allow several sociodemographic control variables education, income, sex, age, race (a dummy variable for Whites), region (a dummy variable for residents of the South), and religious affiliation (dummy variables for members of the three largest religious traditions: evangelical Protestants, mainline Protestants, and Catholics) to affect latent secular and political orientations in waves 3 and Because there is no 22 We categorize respondents religious affiliations into religious traditions based on the method described in Green (2007). We do not include these religious dummies 22

24 theoretical reason to expect sociodemographic effects to vary across panel waves, we constrain them to be equal across the two waves. We further constrain the cross-lagged effects between political orientations and secular orientations and the lagged effects of each variable on itself to be equal across panel waves. If, as our model assumes, the relationship between variables is continuous over time, then, with relatively equal spacing between panel waves, the cross-lagged and lagged effects should be equal across waves (Finkel 1995). Table 4 displays the estimates of all the structural parameters in our models. 23 Not surprisingly, when we correct for measurement error, each secular and political orientation is highly stable over time; stability coefficients all are.86 or greater. [Table 4 Here] Despite this impressive stability, lagged political orientations have statistically significant effects on change in all three secular orientations. First, as confirmation of the dissonance hypothesis, stronger identification with the Democratic Party and more-liberal cultural attitudes are both related to an increased likelihood of identification as None. Liberal ideology has a similar, though in the model of identification as Nones. All sociodemographic variables are measured in wave 2 of the panel. 23 The Supporting Information includes all model estimates, including the confirmatory factor loadings and effects of sociodemographic variables, for the models shown in Tables 4 and 5. 23

25 not quite significant, effect. However, identification as None has no effect on change over time in any of our political dimensions. Thus, being a None seems to be endogenous to politics, but not a mover of political proclivities. Next, we find clear support for the polarization hypothesis. Democratic partisanship, liberal ideology, and liberal cultural attitudes all spur increases over time in passive and active secularism. Unlike identification as a None, passive and active secularism sometimes reciprocate and spur increases in Democratic partisanship and political liberalism. Passive secularism causes increases in liberal ideology and liberal cultural attitudes. Active secularism has a clearly significant effect on cultural attitudes and effects on partisanship and ideology that approach statistical significance. What about the magnitude of our cross-lagged effects? Because the observed indicators of political and secular characteristics are coded to range from zero to one, the unstandardized coefficients represent the impact of an increase in one orientation from its theoretical minimum to its theoretical maximum on change in the other orientation on the same zero-to-one scale. For example, moving from strong Republican to strong Democrat in party identification produces increases of.02 in both passive secularism and active secularism. In more substantive terms,.02 represents about one-sixth of the distance between categories on a zero-to-one scale of worship attendance e.g., between attending two or three times a month versus only once a month. So, it would take about six panel waves or approximately four and one-half years (given the roughly ninemonth average gap between our panel waves) for the average strong Democrat to 24

26 become one category less likely than the average strong Republican to attend worship services. Taking another example, moving from the lowest level of passive secularism to the highest level is associated with an increase of.024 in liberal identification, while the same change in active secularism is associated with an increase of.038 in liberalism. Substantively, that means that it would take about four panel waves (or approximately three years) for the most passively secular respondent to move onetenth of the ideology scale in a more liberal direction than the least passively secular respondent. Meanwhile, it would take just under three panel waves (or a bit more than two years) for the most actively secular respondent to grow more liberal than the least actively secular respondent by one-tenth of the ideology scale. These effects are admittedly modest. However, with corrections for measurement error and a short period between each panel wave, we would not expect them to be large. Over a period of years, the cumulative changes in political orientations based on secularism and in secularism based on politics could be sizeable. 24 As a final step, we evaluate the perceptual hypothesis that both dissonance and polarization apply primarily to people who perceive that religion has infused American politics, particularly the Republican Party. Our test employs two questions that appeared in wave 1 of the SAS panel. One question asked In general, how 24 In the Supporting Information, we also compare the estimated change in one latent variable when we move the other latent variable across what Mplus estimates as its full empirical range. 25

27 much talk about religion is there in politics today?, with three response options: a lot, a little, and no talk about religion at all. The other asked respondents to assess the partisan ties of evangelical Christians, prompting them to say whether evangelicals are mainly Democrats, mainly Republicans, or a pretty even mix of both. Next, we estimated our models of cross-lagged effects between party identification and secular orientations simultaneously for the three groups defined by each of these questions, but allowing the structural parameters to vary across the groups. 25 We expect the effect of partisanship on change in secularism to be strongest for people who see evangelicals as mainly Republicans and who perceive a lot of talk about religion in politics The factor loadings for our latent variables and the effects of demographic variables on the latent variables are held equal across the groups. Only the stabilities of secular and political orientations and the cross-lagged effects are allowed to vary across groups. See the Supporting Information for the full set of estimates. 26 We focus only on party identification because it is the central variable in our hypotheses and the evangelical partisanship variable pertains specifically to party ties. 26

28 Table 5 presents the stabilities and cross-lagged effects separately for each dimension of secularism and for each category of our two conditional variables. 27 The results support the perceptual hypothesis. The impact of party identification on change in identification as a None, passive secularism, or active secularism is never statistically significant unless individuals believe that evangelical Christians are mainly Republicans or perceive a lot of talk about religion in politics. However, among individuals who see evangelicals as primarily Republican and perceive a lot of religious talk in the political environment, the effects of partisanship on secular change are nearly always statistically significant. 28 When Americans perceive a close connection between religion and politics and recognize close ties between traditionalist religion and the GOP, Democratic Party identification spurs increases in secularism For the analysis in which the dependent variable is identification as None and the conditional variable is perceived evangelical partisanship, our latent variable model produces a non-positive definite matrix. So, we estimated the model with observed partisanship and observed identification as None. All other analyses in the table involve latent variables, accounting for measurement error. 28 The only exception is when the dependent variable is identification as None and respondents perceive a lot of talk about religion in politics. 29 To assess whether the differences in effects across perceptions of evangelical partisanship and religious talk in politics are statistically significant, we computed Satorra-Bentler scaled tests of the difference in chi-square between the models in 27

29 [Table 5 Here] Conclusion Our analysis of religion and secularism in American politics offers empirical, theoretical, and substantive innovations. Empirically, we differentiate between the absence of religion and active secularism by introducing an index of active secularism, or the affirmative embrace of a secular identity and worldview. We assess the measurement properties of active secularism in other work (Layman et al. n.d.), where we estimate that just under nine percent of Americans are active secularists (while nearly a quarter are passive secularists). 30 Here, we demonstrate its importance for contemporary politics. Theoretically, our finding that political orientations can shape religious and secular orientations has broader implications. In an increasingly polarized America, political identities structure a wide range of non-political social choices. the table and models that constrain the stabilities and cross-lagged effects to be equal across groups. Though most of the tests do not reach standard levels of statistical significance, they generally suggest that the former set of models fit the data better than the latter. When the conditional variable is perceived evangelical partisanship, the value of the scaled difference in chi-square is 9.74 (p=.28) for identification as none, 5.94 (p=.65) for passive secularism, and (p=.13) for active secularism. When the conditional variable is perceived level of religious talk in politics, the value is (p=.11) for identification as none, (p=.14) for passive secularism, and (p=.01) for active secularism. 30 Part 1 of the Supporting Information has more details. 28

30 Substantively, while past research infers that the intertwining of religion and the GOP has triggered religious non-affiliation among Democrats, we explicitly test the causal mechanism leading to this effect. In the experimental and panel data, the close association of religion and the Republican Party creates cognitive dissonance among Democrats. Many Democrats resolve the dissonance by becoming Nones. Further, the mingling of religion and partisan politics leads to polarization, as Republicans and conservatives grow increasingly religious and Democrats and liberals become more passively and actively secular. Importantly, these processes take shape only when voters perceive the mixture of religion and politics, particularly in the GOP the causal mechanism proposed, but to date untested, in the literature. It is striking that political orientations structure all three forms of secularism: identification as a None, passive secularism and active secularism. The reciprocal relationship between secularism, measured in different ways, and political orientations, also measured in different ways, affirms that this is an important but largely unrecognized cleavage in American politics. It is particularly significant that active secularism plays a role often attributed to religiosity the effect sizes suggest it has a stronger effect on political orientations than they have on it. Thus, people with an actively secular worldview are increasingly found among Democratic identifiers, ideological liberals, and those with left-leaning cultural attitudes. Other research demonstrates that actively secular people are often highly engaged in political activity and that many Democratic Party activists hold actively secular views (Layman and Weaver 2016). 29

31 This historical moment thus resembles the emergence of cultural conservatives led by evangelical Protestants as a political force in the 1970s and 1980s (Layman 2001; Wilcox and Robinson 2010). While secularists may or may not create the same sort of organizational infrastructure as the Christian Right, it is likely that they will increasingly make their voices heard in the political arena. Given the ongoing politicization of religion and secularism we anticipate a continuation of cultural conflict in American politics. 30

32 Bibliography Beard, T. Randolph et al Secularism, Religion, and Political Choice in the United States. Politics and Religion 6(4): Bishop, Bill The Big Sort: Why the Clustering of Like-Minded America Is Tearing Us Apart. 1st Mariner Books ed. Boston: Mariner Books. Bollen, Kenneth A Structural Equations with Latent Variables. New York, NY: Wiley. Campbell, David E., John C. Green, and Geoffrey C. Layman The Party Faithful: Partisan Image, Candidate Religion, and the Electoral Impact of Party Identification. American Journal of Political Science 55(1): Conover, Pamela Johnston, and Stanley Feldman The Origins and Meaning of Liberal/Conservative Self-Identifications. American Journal of Political Science 25(4): Converse, Philip E The Nature of Belief Systems in Mass Publics. In Ideology and Discontent, ed. David Apter. New York: Free Press, Converse, Philip E., and Gregory B. Markus Plus Ça Change : The New CPS Election Study Panel. American Political Science Review 73(1): D Antonio, William V, Steven A. Tuch, and Josiah R. Baker Religion, Politics, and Polarization: How Religiopolitical Conflict Is Changing Congress and American Democracy. Lanham, MD: Rowman & Littlefield. Festinger, Leon A Theory of Cognitive Dissonance. Evanston, IL: Row, Peterson. Finkel, Steven E Causal Analysis with Panel Data. Thousand Oaks, CA: Sage Publications. Goren, Paul, and Christopher Chapp. forthcoming. Moral Power: How Public Opinion on Culture War Issues Shapes Partisan Predispositions and Religious Orientations. American Political Science Review. Green, John C The Faith Factor: How Religion Influences American Elections. Westport, CT: Greenwood Publishing Group. Hansen, Susan B Religion and Reaction: The Secular Political Challenge to the Religious Right. Lanham, MD: Rowman & Littlefield. Hout, Michael, and Claude S. Fischer Why More Americans Have No Religious Preference: Politics and Generations. American Sociological Review 67(2):

33 Explaining Why More Americans Have No Religious Preference: Political Backlash and Generational Succession, Sociological Science 1(October): Iyengar, Shanto, Gaurav Sood, and Yphtach Lelkes Affect, Not Ideology: A Social Identity Perspective on Polarization. Public Opinion Quarterly 76(3): Iyengar, Shanto, and Sean J. Westwood Fear and Loathing across Party Lines: New Evidence on Group Polarization. American Journal of Political Science 59(3): Kellstedt, Lyman A., John C. Green, James L. Guth, and Corwin E. Smidt Grasping the Essentials: The Social Embodiment of Religion and Political Behavior. In Religion and the Culture Wars, Lanham, MD: Rowman & Littlefield, Killan, Mitchell, and Clyde Wilcox Do Abortion Attitudes Lead to Party Switching? Political Research Quarterly 61(4): Kline, Rex B Principles and Practice of Structural Equation Modeling. Fourth edition. New York, NY: The Guilford Press. Layman, Geoffrey C The Great Divide: Religious and Cultural Conflict in American Party Politics. New York, NY: Columbia University Press. Layman, Geoffrey C., David E. Campbell, John C. Green, and Nathanael Sumaktoyo. n.d. Active Secularism and the Politics of Irreligion in the United States. Layman, Geoffrey C., and Christopher Weaver Religion and Secularism among American Party Activists. Politics and Religion 9(2): Leege, David C., Kenneth D. Wald, Brian S. Krueger, and Paul D. Mueller The Politics of Cultural Differences: Social Change and Voter Mobilization Strategies in the Post-New Deal Period. Princeton, NJ: Princeton University Press. Lim, Chaeyoon, Carol Ann MacGregor, and Robert D. Putnam. 2010a. Secular and Liminal: Discovering Heterogeneity Among Religious Nones. Journal for the Scientific Study of Religion 49(4): b. Secular and Liminal: Discovering Heterogeneity Among Religious Nones. Journal for the Scientific Study of Religion 49(4): Margolis, Michele. forthcoming. Cognitive Dissonance, Elections, and Religion: How Partisanship and the Political Landscape Shape Religious Behaviors. Public Opinion Quarterly. 32

34 Mason, Lilliana A Cross-Cutting Calm: How Social Sorting Drives Affective Polarization. Public Opinion Quarterly 80(S1): Patrikios, Stratos American Republican Religion? Political Behavior 30(3): Pierson, Paul Increasing Returns, Path Dependence, and the Study of Politics. The American Political Science Review 94(2): Putnam, Robert D., and David E. Campbell American Grace: How Religion Divides and Unites Us. New York, NY: Simon and Schuster. Satorra, Albert, and Peter M. Bentler Corrections to Test Statistics and Standard Errors in Covariance Structure Analysis. In Latent Variables Analysis: Applications to Developmental Research, Thousand Oaks, CA: Sage Publications, Wilcox, Clyde, and Carin Robinson Onward Christian Soldiers?: The Religious Right in American Politics. Boulder, CO: Westview Press. Wiley, David E., and James A. Wiley The Estimation of Measurement Error in Panel Data. American Sociological Review 35(1):

35 Table 1: Confirmatory Factor Analyses of Passive and Active Secularism Two-Factor Model One-Factor Model Passive Factor Active Factor Indicators Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Passive Secularism Religious attendance Religious guidance Frequency of prayer Belief in God View of the Bible Active Secularism Factual evidence source of true beliefs Great works best source of truth Hard to live based on reason alone Free minds from old traditions/beliefs Values more important than evidence Non-religious guidance Secular identity Correlation between latent factors.74 Goodness of Fit χ 2 (df) χ 2 scaling correction factor CFI RMSEA (53) (52) Satorra-Bentler difference in χ 2 (df) (1) N = 1,909 Note: Coefficients are unstandardized maximum likelihood coefficients. 34

36 Personal Statement Table 2. Variations in the Experimental Treatments No Religion Moderate Religion High Religion Republican My deep roots in the area will help me represent the values of this community. My deep roots in the area and my religious faith will help me represent the values of this community. My deep roots in the area and my commitment to Christ will help me represent the values of this community. Democrat Being a lifelong resident of Martin County will help me to do right by the people of this district. Being a lifelong resident of Martin County and a man of religious faith will help me to do right by the people of this district. Being a lifelong resident of Martin County and a devoted Christian will help me to do right by the people of this district. Issues Republican Democrat Americans are a free people. Government must stop over-spending and threatening that freedom. Society has historically said that marriage is between one man and one woman. To change that definition puts the institution of marriage at risk. Government must continue to provide crucial help for the disadvantaged. We should always help those in need. We need to stop discriminating against gay and lesbian Americans and give them the right to marry the person they love. -- same as no religion The more I pray and read the Bible, the more I know that God has made Americans a free and faithful people. Government must stop over-spending and threatening that freedom. God says marriage should be between one man and one woman. To change that definition puts the institution of marriage at risk. -- same as no religion Government must continue to provide crucial help for the disadvantaged. The Bible says that we should always help those in need. We are all God s children. We need to stop discriminating against gay and lesbian Americans and give them the right to marry the person they love. 35

37 Endorsements Republican Martin County Chamber of Commerce Summerville Independent Business Association Central State Taxpayers Association Martin County Realtors Association Martin County Chamber of Commerce Summerville Independent Business Association Christians for the Traditional Family Martin County Realtors Association Christian Freedom Council Summerville Independent Business Association Christians for the Traditional Family Martin County Realtors Association Democrat Central State Teachers Association Summerville Firefighters Association Martin County Sanitation Workers Martin County Hope Foundation Central State Teachers Association Summerville Firefighters Association Christians for the Common Good Martin County Hope Foundation Faith Alliance for Equality Summerville Firefighters Association Christians for the Common Good Martin County Hope Foundation Memberships Republican Martin County Development Association Summerville Rotary Club United Way Summerville Jaycees Martin County Development Association Oak Street Christian Fellowship Summerville Rotary Club United Way Oak Street Christian Fellowship Brothers of the Cross Christian Community Summerville Rotary Club United Way Democrat Summerville Education Foundation Summerville Kiwanis Club March of Dimes Central State Achievement Society Summerville Education Foundation East Side Christian Church Summerville Kiwanis Club March of Dimes East Side Christian Church Christian Men in Mission Summerville Kiwanis Club March of Dimes 36

38 Photos Republican Democrat 37